Establishment of a radiomics laboratory: data mining and quantitative imaging techniques in clinical radiation oncology

This thesis focuses on the field of data mining and radiomics and its application in clinical radiation oncology from cancer diagnosis to therapies. One of the thesis objective is to establish a "Romagna Imaging Biobank" and apply radiomics to specific oncological pathologies. The goals are to establish a link between tumor phenotype and quantitative image descriptors and personalize anti-cancer therapies. First the thesis outlines a software platform development and a multicentre study for radiomic tools testing, investigating their reproducibility, sensitivity and stability in terms of features extraction. We examine the use of imaging biomarkers, both through phantom-based testing and patient-focused studies: In the first study, software packages compliant with the Image Biomarkers Standardization Initiative (IBSI) were assessed, revealing high standardization in feature implementation. Then we present a multicentre evaluation of dosiomics features, emphasizing their stability and effectiveness in distinguishing dose distributions across various radiation therapy technologies and techniques. The former investigates the use of MR-based quantitative analysis to differentiate testicular tumors, while the latter assesses the complementary roles of MRI-ADC and [68Ga]Ga-PSMA-11-based quantitative analysis in distinguishing prostate cancer patients of varying risk levels. The last patient-focused study introduces a radiomic signature for the classification of benign lipid-poor adrenal adenomas, with potential applications in reducing unnecessary investigations and enhancing the accuracy of not-enhanced CT scans. A review on radiomic studies for lipid-poor adrenal adenomas is in preparation. In conclusion, the outlook of the radiomics laboratory is rooted in the ever-evolving field of data-driven medicine. Collaboration with various medical physics associations and research initiatives, including artificial intelligence, promises to drive progress in clinical oncology. The integration of digital image flows within hospitals and the implementation of standardized practices further enhance a comprehensive cancer care network and strengthen the establishment of a “Romagna Imaging Biobank”